AI, artists, Twitter, oh my
September 7, 2022 9:16 AM   Subscribe

I Went Viral in the Bad Way. Charlie Wurzel talks about AI-created art, the ethical implications he hadn't thought about, and what it's like to become Twitter's Main Character of the Day (read: yelled at by everyone) because of that.
posted by pelvicsorcery (60 comments total) 15 users marked this as a favorite
 
I didn't want to editorialize in the post itself, so I'll do it in a comment: I think he's far harder on himself, and far more generous to his critics, than I would be given his situation. Even though I agree with the larger point his critics are making!

I've never DMed someone to call them a piece of shit. Ever. Not even to Ted Cruz!
posted by pelvicsorcery at 9:17 AM on September 7, 2022 [11 favorites]


There does need to be zero tolerance for AI art (at least, in the automated plagiarism form that it's currently being made in- it's not like procedural generation is a new concept) but there's also no excuse for the dogpiling and screaming. Unpleasant situation all around.
posted by Pope Guilty at 9:32 AM on September 7, 2022 [2 favorites]


Not even to Ted Cruz!

If not me, who? If not now, when?
posted by star gentle uterus at 9:33 AM on September 7, 2022 [13 favorites]


As a musician who has seen the economic viability of audio recordings nosedive as a result of technology, it's not such a surprise to see that illustrators are staring down the barrel of a similar gun.
posted by slkinsey at 9:33 AM on September 7, 2022 [7 favorites]


I think he's far harder on himself, and far more generous to his critics, than I would be given his situation.

Me too. I'm all for hiring and paying artists. But "artists as a class are entitled to illustration contracts for every Atlantic newsletter" is a nonsense position. How would that work? Is there some collective fund all artists share in equally or something?

There's definitely a bunch of complex IP issues around the current crop of deep learning / GAN art and what property rights are appropriate given how these systems are trained on existing images. (Metafilter's own) Andy Baio recently wrote a good pair of articles summarizing where we are with this discussion: Opening the Pandora’s Box of AI Art and Exploring 12 Million of the 2.3 Billion Images Used to Train Stable Diffusion’s Image Generator". Also relevant, the recent Art Made With Artificial Intelligence Wins at State Fair.
posted by Nelson at 9:45 AM on September 7, 2022 [8 favorites]


Photographers are also artists, and if AI art replaces Getty subscriptions it reduces Getty's resources and motivation to pay photographers to keep expanding its library. If he used AI art when he would have otherwise used Getty, then yeah, that's money out of somebody's pocket, even if not a lot.
posted by Pope Guilty at 9:57 AM on September 7, 2022 [10 favorites]


Hmmmm.... personally, I don't like the idea of AI art. However, I have to wonder how many writers and others might simply have put something like "digital art by author" beside that illustration and just gotten on with their lives? How many more will do that now that they've seen this?

I'm not sure this kind of pile-on achieves what the people piling on think it does.
posted by rpfields at 10:15 AM on September 7, 2022 [5 favorites]


More than that is the power of normalization, and the influence that a major publication, especially one that's highly regarded, over what the world and the publishing world thinks of as normalized. Perhaps some might argue that this kind of thing is inevitable; I don't agree, but maybe it's likely. Even if so, there needs to be pushback as long as possible, to give culture, laws, and understanding a chance to develop.
posted by amtho at 10:19 AM on September 7, 2022 [6 favorites]


AI art makes me feel depressed, and I've been struggling to stop worrying and love the bomb since April, and the article covers some of the reasons why. However, there is some schadenfreude to be here. The fact that he used Midjourney to replace a Getty image is particularly delicious, because Getty Images is unambiguously villainous in all the usual monopolistic ways (lowered payouts is one, note that the choice of stock agencies for the author was basically Getty or nothing) but they have a history of going above and beyond by stealing images from the public domain. I'll get out my tiniest violin for Getty Images.
posted by surlyben at 10:25 AM on September 7, 2022 [12 favorites]


I'm actually really surprised it took this long for one of these incidents to crop up.

I've spent about the last ten to fifteen years in newsrooms, through most of the rise of Facebook and Twitter, and one of the deeply under-examined and insidious influences that they've had is the "social card," and specifically the big photo/graphic card. You used to buy a newspaper, or read a news site, and if the article didn't need an image at the top, it wouldn't have one. But now if you share something on social media and it doesn't have a picture with it, you're pretty much guaranteeing that it won't have any reach. As a result, it's been a tremendous boon to services like Getty that can furnish the (e.g. from the article) million vaguely similar pictures of someone like Alex Jones that you need to go with your stories.

My whole job is creating visual stories for news, I've worked with some great photographers and illustrators, I strongly believe people should be paid for creative work, and I was shocked at how quickly after seeing the latest crop of AI art that the darkest portions of my brain coughed up: "you know, we could pipe a field in our CMS into one of these things and immediately free up a bunch of time and budget for other things."

I would not be surprised to see it show up as a feature from the big site vendors within a couple of years. I think that would be really sad and painful, and there's deep moral and ethical problems with it, but the market forces at work here feel pretty unrelenting.
posted by Four String Riot at 10:33 AM on September 7, 2022 [17 favorites]


All of this generative work - code, art, whatever - only works for some definition of "works" for one generation, the one where it's novel. Once it's widespread enough - once it's normalized and being fed back into itself as learning material - the generalized descent of the field of generative media into nothing more than spigot for lukewarm semiotic pablum will be swift.
posted by mhoye at 10:38 AM on September 7, 2022 [8 favorites]


One area where I've been seeing AI art get vigorously discussed is RPG forums. I believe they even banned AI generated art on the D&D subreddit. Considering how big a part artwork plays in RPG supplements vs. the small size of many creators I imagine this is going to be messy for a while.

Just as a personal observation, a lot of one-person-and-their-computer PDF only supplements are filled with public domain art. They aren't paying artists anyways so I imagine there isn't any loss there. However I've been seeing a rash of people selling "packs" of AI generated artwork for RPGs and that seems... bad.
posted by charred husk at 10:44 AM on September 7, 2022 [4 favorites]


Yeah, there's a lot of discussion about this in rpg.net, too. For a lot of authors there, it's a choice between public domain/AI art, and no art at all. Others have mentioned that after the artist and storefront take their cut, there's not much left for the publisher and author.
An issue that's barely been touched on is that if you want specific people, say the same characters, in multiple pieces of art, then things become really difficult. There are also problems with people interacting, like in a fight. The AIs also have no concept of anatomy, so I'll see something that's wonderfully photorealistic, only the subject will have a torn lip, or horror fingers, or one other thing that's really wrong.
posted by Spike Glee at 11:00 AM on September 7, 2022 [1 favorite]


Hmmmm.... personally, I don't like the idea of AI art. However, I have to wonder how many writers and others might simply have put something like "digital art by author" beside that illustration and just gotten on with their lives? How many more will do that now that they've seen this?

"Digital art by author" would involve the creation of art by the author. Charlie did not create anything. He gave a prompt to an algorithm which used that prompt, and millions of images that actually created by artists, to generate an image built out of those images, guided by the prompt. He created nothing.
posted by Pope Guilty at 11:07 AM on September 7, 2022 [5 favorites]


The prompt is creation, the selection is creation. Always has been. That a good portion of labor has been eliminated does not change that.
posted by tychotesla at 11:13 AM on September 7, 2022 [24 favorites]




The prompt is creation, the selection is creation. Always has been. That a good portion of labor has been eliminated does not change that.

All but the least fussy of clients I've worked with have spent far more words on describing what they want than these AI art prompts use. And they select choices at every stage of the process.

If you think a tool can replace artists, then you should at least credit it as one.
posted by Zalzidrax at 11:37 AM on September 7, 2022 [3 favorites]


The prompt is creation, the selection is creation. Always has been. That a good portion of labor has been eliminated does not change that.

Yeah, the part that's actually work, that's actually skilled, that people make their actual livelihood from, that part's been stolen so that people can feel like they're doing art without any of the actual time and effort and creativity and skill that goes into actually creating something, and in a way that ensures that there will over time be less art because the raw materials the algorithms cannibalize to build their output from had to actually be made by somebody and if this algorithmic crap is competing on cost with those somebodies, it's going to win every time.

I'm not even against the technology as a toy, or even as a creative endeavor, but the idea that there's thousands or millions of artists who created works, then there's a group of people who built the algorithm using those works, and there's one guy at the end who typed in a prompt and chose one of the output images, and we're pointing at that one guy at the end and saying "he made this!", that's not only ridiculous but insulting and dangerous. At an absolute minimum the algorithm creators should be forced to acknowledge and compensate the artists whose work they're using, and they should be included in any credits attributed for a given output. They're working people, and we already use technology to fuck workers out of their time, money, and credit enough as it is.
posted by Pope Guilty at 11:44 AM on September 7, 2022 [12 favorites]


I have a feeling a lot of arguments against AI art are going to look like arguments against Dada back in the day.

As to how its made, we aren't asking human artists to list, credit, and recompense the art and artists who have impacted or inspired their own work? And this next part sucks! Living things were the artists before now, gathering practice and inspiration from their life to make art. AI does that but with a few billions images to base its own work off of. And it makes dozens of attempts in the time it takes for a human to pull the canvas off the floor. Even if 99 out of hundred don't get at what the person inputting the prompt wanted that's still 4 logs better than the millionth monkey typing a Shakespeare play by accident.

It's still making something, none of these images are copied from existing ones right?
posted by Slackermagee at 11:45 AM on September 7, 2022 [4 favorites]


There is a blog I stopped reading because (aside from the general decline of the content) they decided to accompany every. single. post. with some AI generated art. It usually looked like it was generated based on the tags of the article, and it just added a lot of visual garbage and noise to their posts. Again, their post content was on the decline anyways, with lots of sponsored posts of cheap shopify junk, but that was the push that made me finally remove it from my Feedly.
posted by msbutah at 11:48 AM on September 7, 2022 [1 favorite]


It's not "making" anything. It's math. It can't think or create. It's not "inspired". It can only do extremely complicated remixing.
posted by Pope Guilty at 11:48 AM on September 7, 2022 [3 favorites]


Pope Guilty, I think people get your point.

(Also, remixing is absolutely art, which rather undermines said point. Remixing has nothing to do with whether a machine can be said to be "creating" here.)
posted by sagc at 11:55 AM on September 7, 2022 [8 favorites]


You're right, I shouldn't have used the term so sloppily.
posted by Pope Guilty at 11:57 AM on September 7, 2022


Extremely complicated remixing is 99.99% of art? We take the concepts and techniques we can get our hands on and wrap our heads around and we make art. Huge leaps and de novo anythings are very rare, iirc, and appear to be limited to the last hudred or two hundred years (its been 14 years since my last art history class).

Humanity didn't leap from inked urn etchings to impressionism, it was thousands of years of very very slow iterative steps done by a small minority of artists. That's something the program will not be able to do as is.
posted by Slackermagee at 11:59 AM on September 7, 2022 [2 favorites]


That said, I do agree that searching and selecting and finding beautiful things in the noise of the universe is very human and creative and artistic.

While I do make art, and do commissions here and there, I don't plan on it being a major source of income for me. The reason I do illustrations is to try and take things in my imagination and get them out in the world in something visible or tangible. And it brings me great happiness that I can do this reasonably well now.

My discomfort with AI art comes less that it will replace artists in a capitalist sense, but in that it might become a seductive and inferior substitute for human imagination.

Art is communication. Often very information-rich communication, most of which is not on the overt and explicit level. Every choice of detail conveys something, intentionally or no, it may be chosen to expression an emotion, or aesthetic, or may just accidentally expose the limits of an artist's knowledge. But with AI art, what information is truly being added by the person working with the tool? A short description. Some broad tweaking that changes the piece entirely. Choosing between a few general packages.

What I fear is that these AI images fill in all this missing information, this would be communication with averages and noise. It generates pieces filled with very intricate looking details that provide very little communication between creator and audience, for all the complexity they seem to have at first glance.

I mean, honestly I'm not so precious to believe that a lot of human artists are all that great at expressing things in detail, that we don't just throw in our own mental average of what we think should be there and get on with things. But sometimes, you realize that there is a choice to be made, and it means something. And very great artists see these choices in so many things. I worry that a great feal is lost when you aren't having to confront the vagueness of your own idea and realize how little form it truly has, then do the work of giving it that form.
posted by Zalzidrax at 12:02 PM on September 7, 2022 [7 favorites]


"Digital art by author" would involve the creation of art by the author. Charlie did not create anything. He gave a prompt to an algorithm which used that prompt, and millions of images that actually created by artists, to generate an image built out of those images, guided by the prompt. He created nothing.

I am sympathetic to the way that technology can steamroll creators, but I'm also struggling with this line of logic, because it's been deployed against other artists who have created complex art that uses technology to re-contextualize previous creations, including DJs and producers sampling music. And I'm not sure it was a correct criticism then, even if it was popular and correct-feeling - though, of course, this benefits from the legitimacy conferred by time passing.
posted by entropone at 12:19 PM on September 7, 2022 [9 favorites]


What bothers me the most about these tools is the complete lack of any attention to the creators of the source images. All of the art that went into training these models is not only not paid for, it's not even credited. I've seen some researchers claim that due to the way the Web is scraped for these images, it's not even possible to reliably credit the artists.

While I've seen arguments that model training is "fair use" in some sense, I've yet to be convinced -- I think it's more that it falls into a weird legal gap in use that we don't yet understand.

If some Google-sized company had spent several billion dollars to purchase or create a large collection of varied images, and then trained their model on that, I think I'd have a less strong reaction to this. I'd still have concerns about these tools displacing human workers, but it would seem like "just" a disruptive new tool, like a super-powered PhotoShop. Rather than representing an outright theft of millions of humans' creativity.
posted by learning from frequent failure at 12:23 PM on September 7, 2022 [9 favorites]


I feel like you can simultaneously:
  1. find what this technology is capable of really neat
  2. find the implications of its impact on working artists disturbing
  3. wish we lived in a society where "making a living" wasn't the tyrant dictating all of our lives
  4. think that this kind of technology won't ruin the actual creative process for anybody, in the same way that photography didn't "ruin" painting,

    AND

  5. find it impossible to divorce the impressive achievements of this AI from the unabashed pilfering of collective cultural output, or notice the way that the AI designers are themselves actively trying to pull of that exact divorce.
I've done some language modeling trained on small data sets. It's great for a joke, because the model doesn't get "smart" until you've trained it on a metric fuckload of data. At that point, what it's doing is more-or-less the equivalent of when people overlay 10,000 photos of faces on top of each other to generate the "perfectly average" face. It just happens to be devising a route through different territory than "average."

As a technological achievement, it's neat! But it's also literally impossible without letting the machine ingest literal libraries worth of knowledge. And opening this up for "artistic production" is one massive mechanical Turk operation, where the machine hides just how much human effort went into its "astonishing feat."

I think it's super cool and bet that artists will think of all sorts of cool things to do with it, almost all of which will involve making it do weird shit rather than the trite regurgitations (which feel more like a party trick to me, if that party trick was capable of destroying labor). But I feel like it's cool for very different reasons than what gets touted. This isn't a computer thinking, no matter how many steampunk Googlers seem to think otherwise. It's a computer mapping out a route through a bunch of invisible human production, as its creators pretend tens of millions of humans aren't actively involved in every output.

(And as a result, it picks up a bunch of dumb human biases too. Andy Baio's exploration of Stable Diffusion's reference images is generally really neat, but the line that stuck out to me is the throwaway one about how the AI was asked to rank images by "aesthetic score," which Baio notes is "largely watercolor landscapes and portraits of women." 🙃)
posted by Tom Hanks Cannot Be Trusted at 12:36 PM on September 7, 2022 [10 favorites]


I am sympathetic to the way that technology can steamroll creators, but I'm also struggling with this line of logic, because it's been deployed against other artists who have created complex art that uses technology to re-contextualize previous creations, including DJs and producers sampling music. And I'm not sure it was a correct criticism then, even if it was popular and correct-feeling - though, of course, this benefits from the legitimacy conferred by time passing.

The vast majority of sampling is about recontextualizing the sample into a new environment/work, which generally doesn't involve attempting to pretend you're not sampling- the Amen break, for example, is immediately recognizable, because the point of using it is not to bury it and pretend you are making something wholly new. You're not saying "this is my thing that I made, on my own"; you're integrating something you didn't make into something you did as part of a larger cultural context and conversation. It's a deliberate and creative act, undertaken by human agency, and samples are chosen carefully by artists to achieve the desired effect and message. This is automated, algorithmic mulching, with no agency or meaning. They are not comparable.
posted by Pope Guilty at 12:58 PM on September 7, 2022 [7 favorites]


Something I hope people understand about the intellectual theft angle is that these ML models don't contain within them replicas of particular works. Instead they provide answers to millions of questions like "If you were to roll a weighted die on how soft a chromatic transition between pupil and iris there is when a word with an angry vibe is present, what would be your answer?". It learned that by looking at copyrighted images.

I don't see that as a violation of copyright. My brain incorporated the results from literally the same process, that I also learned from copyrighted work.
posted by tychotesla at 1:04 PM on September 7, 2022 [7 favorites]


“This is automated, algorithmic mulching, with no agency or meaning.”

Hold on, there's no reason to bring Thomas Kinkade into this discussion.
posted by Ivan Fyodorovich at 1:12 PM on September 7, 2022 [11 favorites]


There's a distinction between whether something is copyrightable and whether something is "art". A lot of widely accepted art pieces are absolutely not copyrightable, or have images which are not copyrightable, but they still count as art. Examples include Duchamp's "Fountain," and Yves Klein's "The Void" (which is one of my favorite paintings, but the image can't be copyrighted because it is intangible)". Samples in music are another example, where a person might use a sample very artfully but still not gain any copyright over the actual sample because they haven't changed it, they've only changed the context.

There is a long history of art and philosophy on the subject of ideas vs execution, and it's fine to come down on the side that says that ideas are the main thing. You will have rigorous arguments on your side, and plenty of excellent examples, particularly in the realm of 20th century fine art. However, there's a competing outlook that says ideas are easy, and execution is the thing. And that idea has it's own strong arguments. When it comes to commercial art, I'd say that execution generally wins. People pay working artists for their concrete output, mostly. And then along comes these AIs to say "well, actually..." which explains why a lot of commercial artists are getting upset about it.

So, like, the prompt. Is the prompt the art? Maybe. Maybe it can be artful. I am skeptical that the prompt is where the value is, or that writing a prompt gives you ownership of the resulting images. The prompt is almost certainly not the image for copyright purposes. If I tried to register a copyright on a painting using only a description of the painting, I'm pretty sure I would have failed spectacularly to register a copyright on the painting.

If I make a Google Images search, there's a certain art to getting the prompt right, there are flags and options I can add, and I would go so far as to admit the possibility of an "artful" prompt, but I wouldn't claim any ownership of the resulting images, nor would I say that the actual images were the result of my creativity. Possibly the specific search results in aggregate are mine, but I think it's a stretch.

In my opinion prompting an AI is more like an image search, though compared to an image search you have less control, because the same prompt will generally return different results.

As for editing, I don't have much to say. Editing already has whatever value it has (considerable, often unsung), and I don't think machine art particularly changes that. If you select an image from a group, or edit a bunch of images together, it shouldn't matter in terms of artfulness whether those images were AI generated or not.
posted by surlyben at 1:21 PM on September 7, 2022 [2 favorites]


[humans] It's a deliberate and creative act, undertaken by human agency, and samples are chosen carefully by artists to achieve the desired effect and message. [ML] This is automated, algorithmic mulching, with no agency or meaning. They are not comparable.

Algorithmic mulching AFTER an artist's prompt, BEFORE an artists selection. If you remove the artist, then yes there is no agency.

Plenty of humans work mightily with a brush and paint, but without control over the prompt and with no desire for selection the process of painting a house is not recognized as an art. Plenty of artists barely think as they write down a few words. Labor is not a fundamental ingredient to art in this sense. ML is just a tool, an expanding of your memory and automatic abilities.
posted by tychotesla at 1:23 PM on September 7, 2022 [2 favorites]


The person giving the algorithm a prompt and choosing from the output options is not an artist. They are an artist's client.

Plenty of humans work mightily with a brush and paint, but without control over the prompt and with no desire for selection the process of painting a house is not recognized as an art. Plenty of artists barely think as they write down a few words. Labor is not a fundamental ingredient to art in this sense.

Necessary and sufficient are different concepts.

ML is just a tool, an expanding of your memory and automatic abilities.

Artists could absolutely use ML as a tool for art, but that would involve something deeper than typing prompts into black-box algorithmic plagiarism automators.
posted by Pope Guilty at 1:45 PM on September 7, 2022 [2 favorites]


So much of this discussion has strong parallels to computer music when it first started being a thing. Kids with laptops (or sequencers, subtle, and drum machines) just typing in notes, they aren't musicians, and they're putting real musicians out of work, etc.

Thing is, any fool can type in notes, yes. But typing in the right ones to make something good is harder. It requires work, either in the moment or in the development of expertise through sustained practice. Not just anyone can open up a DAW and come up with a banger without spending a long time.

And so it is with these AI tools. Any fool can get them to spit out an image. But the image you want, that looks good, that's novel, that' is much harder. It takes patience, a lot of tinkering, some knowledge of his prompts work in practice with the particular generator, etc - work.

It's just a tool. Yes, it will put some people out of work, just as the drum kit did, but there will still be a market for illustrators, photographers, and artists, just as there still is one for marching band drummers.
posted by Dysk at 1:48 PM on September 7, 2022 [13 favorites]


This is automated, algorithmic mulching, with no agency or meaning. They are not comparable.

If you think there's zero agency or meaning in choosing to use this tool, choosing prompt(s) and choosing which final images to use (and where/how to publish them) then I'm not sure it's worth continuing the conversation in this room until we get aligned on the meaning of those words.

My general feeling, watching this (waves at everything) all go down is that we're going to find out that qualities of illustration which seemed integral to the work are actually, generally, not that valuable or important to most people. It's similar to (with important differences) the switch to digital photography — democratizing photography meant a lot more images were being created (and for money!) that were not great according to traditional aesthetic frameworks. And it turned out … it didn't really matter to the consumers in the end.

James Mtume tells a great story on Questlove's podcast about hearing a song using samples for the first time when driving. He pulled over and puzzled at it for a while because the samples weren't in the same key! He overall got into it pretty quickly (the mailbox money from Juicy I'm sure helped) but plenty of folks at that time had very negative reactions: "the things that matter to us as humans don't matter in this new world!"

The differences here are definitely important — sampling doesn't mean you get a full song with the press of a button of course — but the similarities are also instructive.

I'll also note that in both photography and music sampling, decades down the line some of those things that mattered before matter again! Democratizing photography also meant folks could much more easily get really good at it, and use all the tools to make incredible images; and today sampling is so much a part of production workflows that it's become highly refined in how it's done and how those samples are integrated into the final work. Which is to say I think we have no idea what AI art generation will look like or how it will be used and integrated into artistic practice decades down the line.
posted by wemayfreeze at 1:49 PM on September 7, 2022 [6 favorites]


Crossing the streams, What’s Really Behind Those AI Art Images?: Charlie Warzel talks to Andy Baio and Simon Willison, whose articles I linked above.
posted by Nelson at 2:16 PM on September 7, 2022 [2 favorites]


Another thoughtful resource to read about IP rights issues is Creative Commons on AI. They have articles both about the rights of the creators of images used as input to train an AI and also the IP status of the output of an AI. Both are completely open legal questions. Creative Commons has a long history of working to protect the rights of IP creators while also encouraging a productive cultural commons.
posted by Nelson at 2:29 PM on September 7, 2022 [2 favorites]


The person giving the algorithm a prompt and choosing from the output options is not an artist. They are an artist's client. [---] Artists could absolutely use ML as a tool for art, but that would involve something deeper than typing prompts into black-box algorithmic plagiarism automators.

Black-box, algorithmic, automatic, even 'plagiarism', none of these are dealbreakers at all for what makes art and an artist. The Situationists covered all these a half century ago, thousands more artists have before and since, people above are talking about it in terms of electronic music. People use tools. Having the tool do nearly all the work for you is a non-issue.
Situating artistic value in labor is a way of justifying value, which can be appropriate but says little about if something is art or who the artist is. A better thing to look for is who looks at the context and puts together the necessary information to run whatever process that results in something that makes sense for the context. If that makes people uncomfortable about whether a client is part artist, OK - *shrug*. That's a valid discussion to be had, and different working artists feel different ways about it.

"Something deeper than typing prompts" is entirely a matter of degrees. We can definitely say that controls for models are very crude right now, but that says little about the nature of the beast. Art has already visited the ground of simplicity.
posted by tychotesla at 2:39 PM on September 7, 2022 [5 favorites]


the Amen break, for example, is immediately recognizable, because the point of using it is not to bury it and pretend you are making something wholly new

Often the Amen break is used as an explicit reference to past uses of the Amen break, yes, but if someone were to cut up the Amen break and rearrange it to be unrecognizable as an Amen break - or if someone were to carefully arrange different single drum hits to fit the groove of the Amen break, as the producer Boymerang supposedly did to make the break that is now named after him - many would consider that to be a greater technical and creative achievement. And, historically, recognizable sample sources have very often been kept secret, even if the producer would be happy to share them in theory, because they are on shaky legal ground.

It’s hard to really make an analogy with ML-based art work here, because the uncomfortable thing about ML-based art is that since a human isn’t doing the procedural part of the remix, it becomes difficult to use any of our existing yardsticks to judge what’s meaningfully original.
posted by atoxyl at 5:37 PM on September 7, 2022 [2 favorites]


Personally I have a hard time seeing how these tools are qualitatively different from remixing tools of the past. Rather what we’re grappling with is a new dimension to the old issue that our intellectual property regime ends up favoring the largest players. Heretofore the major reason for this unevenness is unequal distribution of legal resources, but now it is compounded by unequal distribution of computing resources.
posted by atoxyl at 5:46 PM on September 7, 2022


Thanks to everyone who ever wrote/composed/drew/devised/painted/photographed/designed/shared/shooped/invented/performed/gmailed/etc anything online ever. All grist for the mill! Your sacrifice will be remembered (but actually not, soz)
posted by I'm always feeling, Blue at 7:53 PM on September 7, 2022 [1 favorite]


“Go to a website and most of the image content is hosted elsewhere. Articles are full of embedded tweets or Instagram posts or stock photography."

Don't forget that some people share their visual art with various CC and other non-copyright licenses, and that others use them accordingly.
posted by doctornemo at 8:52 PM on September 7, 2022


In my opinion prompting an AI is more like an image search, though compared to an image search you have less control, because the same prompt will generally return different results.

From my limited mucking around with one of the StaleDiffusion (typo, but I'm keeping it) self-contained systems....if you use the same seed and settings, you will get the same image twice from the same prompt.

Thanks to everyone who ever wrote/composed/drew/devised/painted/photographed/designed/shared/shooped/invented/performed/gmailed/etc anything online ever. All grist for the mill! Your sacrifice will be remembered (but actually not, soz)

My brain does exactly this anyway. The AI has just seen more .jpgs than I have. It's not like the AI is taking a copy of the image and filing it away in a database for later plagiarism. It's noting things about it, much like you would if you saw a picture in a gallery, and it uses those (gigabytes of) notes to draw inferences that it later applies to prompts.

From my perspective, there are artistic choices involved in creating a successful picture. 'Just' inputting a prompt isn't going to get you an image worth sharing with the world. You need to refine and develop, sculpting the response, by providing simplified versions of what you really want, or by focussing on successful elements from previous iterations. Perhaps cooking is a better metaphor. There's a lot of flavours out there (inferences drawn from images in the training set), you're only going use a few in a recipe (those codified relationships suggested/filtered by the prompt), and actual recipes, even those based on prior art, often take multiple attempts to perfect. Once you've got the recipe though, it's repeatable. Can a recipe and its result be art?

Sure, I say. My art gallery is a broad tent.

Because I contain multitudes, though, I can also counterargue that AI provides safety rails, that you can get something 'edible' without trying particularly hard. Over time this will only get easier, the first try ever closer to something of artistic merit, - where, then, is the art if an inartistic choice is impossible?

To which I haughtily respond, meh. Art may be worth suffering for, but being suffered for is not intrinsic to art. If we could get great art at the push of a button, so what? Am I arguing that we deserve less great art?

At this point I acuse myself of twisting my own words against me, and must draw a curtain across the resultant unpleasantness.
posted by Sparx at 9:15 PM on September 7, 2022 [4 favorites]


tafilter: to which I haughtily respond, meh.
posted by Ryvar at 9:17 PM on September 7, 2022 [4 favorites]


in the same way that photography didn't "ruin" painting

I haven't given much thought to AI art/commercial ethics thing one way or the other, but here are a couple thoughts that might be relevant.

1. Paintings from before photography seemed like they were more often likely to be photorealistic. So if photography didn't "ruin" painting, I get the impression that it at least changed the expectation that artists would be able to generate true-to-life imagery.

2. I think that in the future we'll see a movement of people who try to avoid the use of AI the way people do veganism or wearing only organic natural fabrics or whatever or bikes instead of cars.
posted by aniola at 1:31 AM on September 8, 2022 [3 favorites]


It's not AI, stop calling it AI.
posted by GallonOfAlan at 2:33 AM on September 8, 2022 [3 favorites]


I think the copyright issue is an interesting one, but ultimately I think the fair use defense is pretty strong in this case - it's pretty transformative. Possibly a better discussion to have is one based in the moral rights of the original artists. The US's conception of moral rights is pretty narrow, but in other jurisdictions that might prove to be a more productive line of attack than copyright per se.
posted by vibratory manner of working at 3:38 AM on September 8, 2022 [3 favorites]


I say again:

Art Theft.

*Art Theft.*

ART THEFT!
posted by Faintdreams at 5:03 AM on September 8, 2022 [1 favorite]


It's not AI, stop calling it AI.

If you have a clear definition of what is and is not AI I’m sure we’d all benefit from it. As it stands the definition is not clear cut at all.
posted by wemayfreeze at 1:00 PM on September 8, 2022 [2 favorites]


Well, it's certainly not AGI. I got a lot less grumpy when I stopped seeing "AI" as "AGI" and accepted "AI" as a marketing term.
posted by Ivan Fyodorovich at 5:35 PM on September 8, 2022 [1 favorite]


I say again:

Art Theft.

*Art Theft.*

ART THEFT!


I have extremely low faith in non-artists to care about art theft since it's not like they ever cared about its other forms, nor would they even recognize it. Most of the threads on this topic remind me of when a certain social media site I used for gaining follows for art commissions had a mass exodus, and my non-artist friends would be like "Have you tried uploading your stuff to 9gag/reddit/imgur/etc? There's lots of images on there!" not understanding the difference between the function of an image aggregate site full of uncredited reposts vs places to get personal followings.

Lookin forward* to the gradual replacement of
"You're an artist? Can you draw me?"
and "Can you draw for the exposure?"
and "Oh you're a DIGITAL artist? But doesn't the computer just do that for you?"

with "Oh, you're an artist? Check it out, me too!! I made this thing in MidJourney. Have you tried it?"


*not looking forward at all
posted by picklenickle at 10:03 PM on September 8, 2022


I suppose artists could start framing it not as "art theft" or "piracy" (which sound cool and hip) and instead point out that their works being used non-consensually to feed paid software is a form of wage theft.

To me, as an artist, discussions of "is it art" or the "ease" of it is not important. Having tools to make art easy and accessible is GREAT. I would love for there to just be something that could just instantly render an image or model directly from my brain and then my creative output would be extremely efficient. Or a way to snap my fingers and make the thing appear. I could make an entire virtual reality game or animated series all by myself if something like that existed-----but I don't like having someone's hard work be put into a grinder without permission for someone else's profit.
posted by picklenickle at 10:23 PM on September 8, 2022


text-to-video coming soon...
posted by gwint at 6:29 AM on September 9, 2022 [1 favorite]


Does it make it more of an art creation if you guide the output with your own drawings?

Though, looking at it merely as a lossy image compression algorithm, compacting several hundred terabytes of images down into a 4.27 gigabyte "weights" file that can get anything resembling them back out of it was previously the realm of science fiction.
posted by fragmede at 3:08 AM on September 10, 2022 [2 favorites]


Where does something (someone?) like Loab fit into this conversation?
posted by rhamphorhynchus at 4:01 AM on September 10, 2022 [1 favorite]


Online Art Communities Begin Banning AI-Generated Images
However you feel about the ethics of AI art, online art communities are facing a very real problem of scale: AI art can be created orders of magnitude faster than traditional human-made art.
posted by Nelson at 8:11 AM on September 10, 2022 [1 favorite]


And so it begins...
Promptbase: DALL·E, GPT-3, Midjourney, Stable Diffusion Prompt Marketplace
Find top prompts, produce better results, save on API costs, sell your own prompts.
posted by thatwhichfalls at 3:12 PM on September 12, 2022 [2 favorites]


This artist is dominating AI-generated art. And he’s not happy about it.
these open-source programs are built by scraping images from the internet, often without permission and proper attribution to artists. As a result, they are raising tricky questions about ethics and copyright. And artists like Rutkowski have had enough. ...

Rutkowski says he doesn’t blame people who use his name as a prompt. For them, “it’s a cool experiment,” he says. “But for me and many other artists, it’s starting to look like a threat to our careers.”
posted by Nelson at 9:56 AM on September 16, 2022


Getty Images bans AI-generated content over fears of legal challenges.
There are real concerns with respect to the copyright of outputs from these models and unaddressed rights issues with respect to the imagery, the image metadata and those individuals contained within the imagery
posted by Nelson at 4:35 PM on September 21, 2022


« Older Twenty years of Languagehat   |   “I can lose a bag of flour to have a good... Newer »


This thread has been archived and is closed to new comments